An Effective Computational Method for Predicting Self-Interacting Proteins Based on VGGNet Convolutional Neural...

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Self-interacting proteins (SIPs) are vital for cellular processes and disease association.
• Experimental identification of SIPs is costly and time-consuming.
• Accurate computational prediction of SIPs remains a significant challenge.

Purpose of the Study:

• To develop a novel computational method for predicting Self-interacting proteins (SIPs).
• To leverage protein sequence data and advanced machine learning techniques for improved SIPs prediction.

Main Methods:

• The VGGNGLCM method integrates VGGNet (VGGN) deep convolutional neural network with Gray-Level Co-occurrence Matrix (GLCM).
• Position Specific Scoring Matrix (PSSM) was used to capture evolutionary information, with features extracted by GLCM.
• VGGNet served as the predictive classifier for identifying Self-interacting proteins.

Main Results:

• The VGGNGLCM model achieved high prediction accuracies: 95.68% for yeast and 97.72% for human datasets.
• VGGNGLCM demonstrated superior performance compared to Convolutional Neural Network (CNN) and Support Vector Machine (SVM) classifiers.
• Experimental validation confirmed the effectiveness and robustness of VGGNGLCM against existing methods.

Conclusions:

• VGGNGLCM is an effective and robust computational tool for predicting Self-interacting proteins (SIPs).
• The method offers high accuracy and can significantly advance bioinformatics research in SIPs prediction.
• VGGNGLCM provides a valuable alternative to experimental methods for identifying protein interactions.